From the comments, on high frequency trading

This is from a high frequency trader, in response to my link to this paper (pdf), favoring batch auctions:

The author’s purported cure is far worse than the disease. Positional externalities from shaving latency are indeed real, but they’re not really that large relative to market size. A good way to estimate their magnitude is by how much money has been spent on cutting down the Chicago-NYC messaging latency, the two most liquid and hence profitable trading centers. The cost recently spent on this infrastructure (largely microwave relay networks) is about $500 million. Assume that the infrastructure depreciates in about a year and generously assume that the spending on intra-market latency is about the same.

That’s a total cost of about $1 billion/yr in market costs imposed by latency based positional externalities. American equity markets trade $24 quadrillion in value a year (and that’s only counting shares, not derivatives). Which means the cost to the typical investor of the latency externalities comes out to an upper bound of $4.5e-05 per dollar traded, or for example to trade one share of MSFT: $0.0016. That’s the upper bound of cost savings by perfectly eliminating latency externalities. The cost certainly isn’t trivial, but it is much lower than the forced imposition of $0.005 in bid-ask trading costs because the SEC refuses to decrease the minimum $0.01 tick size. With an economical tick size the average bid-ask spread would easily go in half. (Plus it would reduce the latency externalities since market makers could price improve rather than rushing to jump first in line the order book queue). My point being is that if we’re that worried about reducing costs to investors there’s an alternative that we’re already ignoring that both has a larger impact and poses much less risk than completely tearing up the foundations of the market structure.

Finally the authors assume that batch auctions don’t come with any of there own structural costs. Not only do they indeed have substantial defects themselves, but they don’t even eliminate the latency externalities. The market already uses batch auctions at market open and close. As any trader will tell you these are far more manipulated than continuous trading. During a batch auction an indicative price is published prior to crossing based on the currently resting buy and sell orders. A trader can easily change this indicative price or imbalance by entering a large order and canceling it before auction. Analogous strategies aren’t impossible, but are much harder in continuous trading because a resting order can be crossed at any time, and hence poses real economic risks to the trader. To paraphrase Alex continuous trading acts as a tax on bullshit.

The flip side of a pre-cross indicative price is that traders will wait for as long as possible before the cross to enter their orders. No trader using proprietary signals is going to want other market participants to see his order for any longer than is absolutely necessary. The counter-strategy being shaving down your latency even further so you get to see others’ orders first. Then modify yours accordingly by trading even closer to the cross time using your lower latency. So what frequently happens in opening and closing batch auctions is that the order book and indicative price is pretty much garbage until a few milliseconds before the cross, at which point the real price formation occurs. When I worked in a much larger HFT firm I was a continuous guy, but sat next to the batch auction guys. We certainly cared about our latency, but generally we focused much more on our signals and execution algorithms. The auction guys in contrast were always obsessed with their latency.

Switching to batch auctions will not reduce the cost of latency positional externalities, and is pretty likely to increase them. On top of all that it will give us a much lower-quality and less efficient market structure. There are certainly better ways to tackle the latency externality costs. But it’s important to recognize the perfect’s the enemy of the good here, I doubt we can ever fully eliminate them under any sane structure. It’s better to think of moderate improvements that work on the margin, rather than centrally planned grand sweeping re-designs of the entire market structure.

At the first link, in the comments, he has several follow-up explanations, all recommended.


Comments for this post are closed